lebsral

Find every way users can break your AI before they do. Use when you need to red-team your AI, test for jailbreaks, find prompt injection vulnerabilities, run adversarial testing, do a safety audit before launch, prove your AI is safe for compliance, stress-test guardrails, or verify your AI holds up against adversarial users. Covers automated attack generation, iterative red-teaming with DSPy, and

See exactly what your AI did on a specific request. Use when you need to debug a wrong answer, trace a specific AI request, profile slow AI pipelines, find which step failed, inspect LM calls, view token usage per request, build audit trails, or understand why a customer got a bad response. Covers DSPy inspection, per-step tracing, OpenTelemetry instrumentation, and trace viewer setup., debug slow

Track which optimization experiment was best. Use when you have run multiple optimization passes, need to compare experiments, want to reproduce past results, need to pick the best prompt configuration, track experiment costs, manage optimization artifacts, decide which optimized program to deploy, or justify your choice to stakeholders. Covers experiment logging, comparison, and promotion to prod

Use when you need to customize how DSPy formats prompts for a specific provider — switching from chat to completion format, forcing JSON output, or debugging prompt rendering issues. Common scenarios - debugging why your prompt looks wrong when sent to the model, switching from OpenAI to Anthropic and the formatting breaks, forcing the model to return valid JSON instead of markdown, working with c

Put your AI behind an API. Use when you need to serve AI features as web endpoints, add AI to an existing backend, deploy AI for other services to call, wrap a DSPy program in REST or HTTP, build an AI microservice, or put a language model behind FastAPI or Flask. Also use for deploy AI model to production, AI REST API, serve DSPy program over HTTP, Docker AI service, AI endpoint for mobile app, h

Analyze images and extract structured data from visual content using AI. Use when analyzing product photos, extracting text from screenshots, generating alt text for accessibility, visual question answering, categorizing images by content, reading receipts and invoices from photos, OCR with AI, describing images for search indexing, product photo categorization, document image processing, chart an

Use when output quality varies across runs and you want to sample multiple completions and pick the best — trading latency for reliability on high-stakes outputs. Common scenarios - generating multiple candidate answers and picking the highest-scoring one, improving reliability on high-stakes classification, reducing variance in creative generation, getting better summaries by sampling several and

Use when you have already tried prompt-only optimization and want the next level — jointly tuning prompts and model weights for maximum quality. Common scenarios - you have maxed out prompt optimization and need the next level, combining instruction tuning with weight tuning for maximum quality, making a small model match a large model through joint optimization, or squeezing the last few percent

Use when you have 50+ labeled examples and want a quick accuracy boost as your first optimization step — the simplest and fastest DSPy optimizer. Common scenarios - your first optimization attempt on a new DSPy program, adding few-shot examples automatically from labeled data, quick accuracy boost before trying heavier optimizers, bootstrapping demonstrations from a teacher model, or getting start

Auto-sort, categorize, or label content using AI. Use when sorting tickets into categories, auto-tagging content, labeling emails, detecting sentiment, routing messages to the right team, triaging support requests, building a spam filter, intent detection, topic classification, or any task where text goes in and a category comes out. Also use when classification accuracy varies between runs or sem

Stop your AI from making things up. Use when your AI hallucinates, fabricates facts, is not grounded in real data, does not cite sources, makes unsupported claims, or you need to verify AI responses against source material. Also use when your LLM makes up facts, responses are disconnected from the input, or outputs are not grounded in source documents. Covers citation enforcement, faithfulness ver

Condense long content into short summaries using AI. Use when summarizing meeting notes, condensing articles, creating executive briefs, extracting action items, generating TL;DRs, creating digests from long threads, summarizing customer conversations, or turning lengthy documents into bullet points. Also used for AI summary too generic, summarize Slack threads, condense customer feedback, meeting

Switch AI providers or models without breaking things. Use when you want to switch from OpenAI to Anthropic, try a cheaper model, stop depending on one vendor, compare models side-by-side, a model update broke your outputs, you need vendor diversification, or you want to migrate to a local model. Also use when your prompt broke after a model update, prompts that work for GPT-4 do not work for Clau

Build AI that takes actions, calls APIs, and does things autonomously. Use when you need AI to call APIs, use tools, perform calculations, search the web and act on results, interact with databases, or do multi-step tasks. Also AI that does things not just talks, tool-using AI agent, AI calls external APIs, function calling with DSPy, build AI that books appointments, AI workflow automation, agent

Translate text between languages with AI while preserving brand voice and terminology. Use when translating app copy to Spanish, localizing marketing content, multilingual support tickets, i18n with AI, machine translation with brand voice, translating product descriptions, localizing help docs, batch translating i18n JSON files, glossary-enforced translation, AI-powered localization pipeline, tra

Generate articles, reports, blog posts, or marketing copy with AI. Use when writing blog posts, creating product descriptions, generating newsletters, drafting reports, producing marketing copy, creating documentation, writing email campaigns, or any task where AI writes long-form content from a topic or brief. Powered by DSPy content generation pipelines., AI blog writer, generate marketing copy

REMOVED IN DSPy 3.x -- use dspy.Refine or dspy.BestOfN instead (see /dspy-refine, /dspy-best-of-n). Legacy documentation for dspy.Assert and dspy.Suggest kept for existing codebases only. For new code, use dspy.Refine (iterative improvement with feedback) or dspy.BestOfN (sampling, pick best). Also used for dspy.Assert, dspy.Suggest, runtime validation for LLM output, retry on bad output, backtrac

Use when you need to run DSPy modules asynchronously — FastAPI endpoints, concurrent LM calls, non-blocking execution, or integrating DSPy into async web frameworks. Common scenarios - serving DSPy behind FastAPI or Starlette, running multiple LM calls concurrently with asyncio.gather, non-blocking batch processing, combining async with streaming, or building async agent loops. Related - ai-servin

Use when you need structured source attribution in AI responses — verifiable citations that link claims to specific passages in source documents. Common scenarios - RAG with citation extraction, grounded answers with document references, legal or compliance use cases requiring source proof, building AI that cites its sources, or verifying which document an answer came from. Related - ai-stopping-h

Use when the agent task is best solved by writing and executing Python code — data manipulation, computation, file processing, or tasks where code is more reliable than natural language reasoning. Common scenarios - data analysis tasks where the agent writes pandas code, computation-heavy tasks where natural language reasoning fails, file processing automation, tasks requiring precise calculations

Use when you need to prepare training/dev data for DSPy optimizers — loading from CSV/JSON/HuggingFace, creating Examples, setting input keys, or building train/dev splits. Common scenarios - loading a CSV of labeled examples for optimization, converting HuggingFace datasets to DSPy format, creating train/dev/test splits, building Examples with proper input keys, converting JSON data for DSPy, or

Use when you have multiple optimized versions of a program and want to combine them — voting, averaging, or routing across program variants for more robust outputs. Common scenarios - you have optimized several versions of a program and want to combine the best ones, using majority voting across multiple programs for higher accuracy, building a robust system by routing to different specialized pro

Use when you need to measure how well your DSPy program performs — writing metrics, scoring against a dev set, or comparing before/after optimization. Common scenarios - measuring accuracy before and after optimization, writing custom metrics for your task, scoring a program against a held-out dev set, comparing two prompt strategies, building a test suite for AI quality, or running regression tes

Use Langtrace for DSPy observability and tracing. Use when you want to set up Langtrace, langtrace-python-sdk, auto-instrument DSPy, trace DSPy calls, LLM observability, app.langtrace.ai, or self-hosted tracing. Also used for langtrace setup, langtrace API key, pip install langtrace-python-sdk, DSPy tracing, auto-instrument DSPy, langtrace self-hosted, langtrace docker, trace LM calls, langtrace v

Use when basic BootstrapFewShot is not enough and you want to search over multiple candidate demo sets — better results at the cost of more LM calls. Common scenarios - BootstrapFewShot alone is not reaching target accuracy, you want to search over multiple candidate demo sets and pick the best, optimizing for tasks where example selection matters a lot, or when you have compute budget for a more

Use when the task benefits from intermediate reasoning before producing an answer — multi-step logic, analysis, math, or complex classification where direct prediction fails. Common scenarios - classification tasks where the model needs to reason about edge cases, math word problems, multi-step analysis, complex question answering, legal or medical reasoning, any task where thinking before answeri

Use when you want to optimize instructions by generating many candidates and picking the best — useful when few-shot demos alone are not enough and you want to tune the task description itself. Common scenarios - your current task instructions produce mediocre results, you want to automatically generate and test many instruction variants, the task is hard to describe in one sentence, or few-shot e

Use when you want to optimize instructions without few-shot examples — a lightweight alternative to COPRO when you do not have or do not want to use demonstrations. Common scenarios - optimizing instructions when you do not have or do not want to use few-shot demonstrations, lightweight instruction search as a first step, tasks where examples in the prompt confuse the model, or when you want fast

Use when you want to extract interpretable decision logic from labeled examples — generating explicit rules that explain patterns in your data. Common scenarios - extracting business rules from labeled classification examples, understanding why a model makes certain predictions, generating human-readable decision criteria from data, building interpretable classifiers, or documenting implicit label

Use when you want few-shot demos that are dynamically selected per input based on similarity — better than fixed demos when inputs vary widely. Common scenarios - inputs vary widely and fixed examples do not cover enough cases, dynamically selecting the most relevant demos per input, building a retrieval-augmented prompt with similar examples, or when static few-shot examples work for some inputs

Use when you have hand-picked high-quality examples and want to use them directly as few-shot demonstrations — no bootstrapping, just your curated demos. Common scenarios - you have expert-curated examples that you trust more than bootstrapped ones, hand-picked demonstrations for high-stakes tasks, using existing labeled data directly without bootstrapping, or when you want full control over which

Use Qdrant as a vector database with DSPy, or connect any vector DB (Pinecone, ChromaDB, Weaviate) with custom retrievers. Use when you want to set up Qdrant, QdrantRM, dspy-qdrant, vector database for DSPy, vector search, hybrid search, or build custom retrievers for Pinecone, ChromaDB, or Weaviate. Also used for qdrant, dspy-qdrant, QdrantRM, vector database, vector search, pinecone DSPy, chroma

Use when you need maximum quality from a smaller/cheaper model — generates training data from a teacher model and fine-tunes a student model weights. Common scenarios - distilling GPT-4 quality into a cheaper model, generating training data from a strong teacher to fine-tune a weak student, reducing inference costs by replacing an expensive model with a fine-tuned small one, or building a producti

Deep dive into dspy.ChatAdapter -- the default adapter that formats DSPy signatures into multi-turn chat messages with field delimiters, parses LM responses back into typed Python objects, and falls back to JSONAdapter on failure. Use when you need to understand how DSPy builds prompts, debug why a model ignores output format, customize prompt rendering, enable native function calling, use callbac

LLM observability for DSPy with Langfuse -- auto-trace every LM call, attach scores and evaluations, run annotation queues for human review, and track experiments across prompt versions. Use when you want to set up Langfuse, langfuse.com, openinference-instrumentation-dspy, trace DSPy calls, LLM observability with scores, annotation queues, or experiment tracking. Also used for langfuse setup, pip

Use when you want higher accuracy by generating multiple reasoning chains and selecting the best answer — trading speed for quality on critical outputs. Common scenarios - high-stakes decisions where you want multiple reasoning paths compared, classification tasks where one chain of thought is not reliable enough, improving accuracy by generating several answers and selecting the best-reasoned one

Use when you have independent LM calls that can run concurrently — batch processing, fan-out patterns, or speeding up pipelines with no data dependencies between steps. Common scenarios - processing a batch of inputs through a DSPy module concurrently, fan-out patterns where multiple independent LM calls run at once, speeding up evaluation by parallelizing predictions, or reducing wall-clock time

Use Arize Phoenix for DSPy tracing and evaluation. Use when you want to set up Phoenix, arize-phoenix, openinference, DSPyInstrumentor, open-source trace viewer, localhost:6006, or LLM evals. Also used for phoenix setup, arize phoenix, pip install arize-phoenix, phoenix local UI, phoenix evaluations, DSPy trace viewer, open-source LLM observability, phoenix vs langtrace, openinference-instrumentat

Use when the mapping from input to output is straightforward and does not need reasoning steps — simple classification, extraction, formatting, or Q&A where minimal latency matters. Common scenarios - simple classification tasks, basic extraction, format conversion, straightforward Q&A, or any task that does not benefit from chain-of-thought reasoning — when you want the fastest possible LM call.

Use when the task requires calling external tools or APIs to gather information — multi-step tool use with reasoning, like searching databases, calling APIs, or combining multiple data sources. Common scenarios - building agents that search the web and synthesize results, multi-step information gathering from APIs, chatbots that look up data before answering, question answering that requires exter

Use LangWatch for DSPy auto-tracing and real-time optimizer progress. Use when you want to set up LangWatch, langwatch.dspy.init, auto-tracing DSPy, real-time optimization dashboard, optimizer progress tracking, app.langwatch.ai, or DSPy optimizer dashboard. Also used for langwatch setup, pip install langwatch, langwatch trace, optimizer progress, real-time optimization, watch optimizer run, LangW

Use when you need to configure which language model DSPy uses — setting up providers, API keys, model parameters, or assigning different models to different pipeline stages. Common scenarios - setting up OpenAI or Anthropic API keys, configuring model parameters like temperature and max_tokens, using different models for different pipeline stages, switching between providers, using local models wi

Use when you need to connect DSPy agents to external MCP tool servers — databases, file systems, APIs, or any MCP-compatible service. Common scenarios - wiring MCP tools into a ReAct or CodeAct agent, discovering tools from an MCP server at runtime, converting MCP tools to DSPy tools, or building agents that use tools hosted on remote servers. Related - dspy-tools, dspy-react, dspy-codeact, ai-tak

Use when you want the highest-quality prompt optimization DSPy offers — jointly optimizes instructions and few-shot demos, with auto=light/medium/heavy presets. Common scenarios - you want the best possible accuracy from prompt optimization, jointly tuning instructions and few-shot demonstrations, using auto presets for different compute budgets, or when COPRO or BootstrapFewShot alone are not rea

Use MLflow for DSPy tracing, experiment tracking, and model registry. Use when you want to set up MLflow, mlflow.dspy.autolog, MLflow Tracing, MLflow experiment tracking, MLflow model registry, or full ML lifecycle management. Also used for mlflow setup, pip install mlflow, mlflow.set_experiment, mlflow UI, mlflow model versioning, mlflow OpenTelemetry, mlflow vs wandb, mlflow tracing DSPy, regist

Use when you need to compose multiple DSPy calls into a pipeline — structuring multi-step programs as reusable, optimizable components with forward() logic. Common scenarios - building a multi-step pipeline as a class, composing Predict and ChainOfThought calls in sequence, creating reusable AI components, structuring a RAG pipeline as a module, or building nested programs where one module calls a

Run DSPy with local models using Ollama — no API key needed. Use when you want to run DSPy locally, use Ollama, set up a local LLM, run offline, or configure local model parameters. Also used for ollama, local model, run LLM locally, llama local, self-hosted LLM, ollama serve, ollama_chat, local inference, run DSPy offline, no API key needed, ollama pull, ollama list, ollama rm, num_ctx, ollama co

DSPy typed wrappers (dspy.Image, dspy.Audio, dspy.Code, dspy.History) for multimodal data in signatures. Use when working with non-text inputs like images, audio, or code, building multimodal AI pipelines, processing images alongside text, handling audio transcription inputs, working with code files as typed inputs, or managing conversation history in multi-turn chatbots. Also used for multimodal